Hello.
I use pi cameras for Aruco markers detection and recently noticed that markers position shift even if scene doesn't change. 1 pixel shift can be explained with sensor noise + Aruco detectoin algorithm, but I observe 3-4 pixels change, sometimes more.
I've decided to verify with next setup:
- 2 cameras on platform, 10 cm from each other, looking in one direction
- 5 markers, ~1 meter away. 4 close to the corners of photo, 1 in center.
- camera.configure(camera.create_preview_configuration(main={"format": "RGB888", "size": (2304, 1296)}))
- camera.set_controls({"FrameRate": 100, "AfMode": controls.AfModeEnum.Manual, "LensPosition": 0, "ExposureTime": 4000})
- making 500 pairs of photos, ~10 photos per second on each camera
- detect markers with openCV and for each marker take average x/y of 4 corners
- plot 500 average points for x/y and left/righ camera (gaussian smoothing)
Some results:
- As shift is different on 2 cameras we can exclude noise from moving platform
- X is more stable, but Y can be up to 7 pixels
1. In last test I noticed that there is a gap in right camera between camera and board. Is it ok if I completely tear it off and glue it again?
2. I noticed that on left camera there is no such gap but still camera can be wiggled relatively to board. This layer between camera and board is soft, what is the reason for this? Is it vibration dampener? Thermal regulation? Does it make sense to use superglue all around it to completely exclude physical movement of camera.
3. There is a small piece of plastic in front around lens, which is easily moved with fingernail? Why is it needed?
4. probably physical movement is not the only reason, I found other post with similar issue (viewtopic.php?t=227982). Could you recommend other Pi cameras, maybe more expensive, which are more stable?
I use pi cameras for Aruco markers detection and recently noticed that markers position shift even if scene doesn't change. 1 pixel shift can be explained with sensor noise + Aruco detectoin algorithm, but I observe 3-4 pixels change, sometimes more.
I've decided to verify with next setup:
- 2 cameras on platform, 10 cm from each other, looking in one direction
- 5 markers, ~1 meter away. 4 close to the corners of photo, 1 in center.
- camera.configure(camera.create_preview_configuration(main={"format": "RGB888", "size": (2304, 1296)}))
- camera.set_controls({"FrameRate": 100, "AfMode": controls.AfModeEnum.Manual, "LensPosition": 0, "ExposureTime": 4000})
- making 500 pairs of photos, ~10 photos per second on each camera
- detect markers with openCV and for each marker take average x/y of 4 corners
- plot 500 average points for x/y and left/righ camera (gaussian smoothing)
Some results:
- As shift is different on 2 cameras we can exclude noise from moving platform
- X is more stable, but Y can be up to 7 pixels
1. In last test I noticed that there is a gap in right camera between camera and board. Is it ok if I completely tear it off and glue it again?
2. I noticed that on left camera there is no such gap but still camera can be wiggled relatively to board. This layer between camera and board is soft, what is the reason for this? Is it vibration dampener? Thermal regulation? Does it make sense to use superglue all around it to completely exclude physical movement of camera.
3. There is a small piece of plastic in front around lens, which is easily moved with fingernail? Why is it needed?
4. probably physical movement is not the only reason, I found other post with similar issue (viewtopic.php?t=227982). Could you recommend other Pi cameras, maybe more expensive, which are more stable?
Statistics: Posted by bendyna — Wed Mar 12, 2025 9:43 pm — Replies 1 — Views 45